Machine learning techniques for predictive prioritization

Patent metadata

Field	Value
Publication number	US-2022147865-A1
Application number	US-202017096062-A
Country	US
Kind code	A1
Filing date	Nov 12, 2020
Priority date	Nov 12, 2020
Publication date	May 12, 2022
Grant date	—

Abstract

Various embodiments of the present invention provide methods, apparatus, systems, computing devices, computing entities, and/or the like for performing predictive prioritization. Certain embodiments utilize systems, methods, and computer program products that perform predictive prioritization using a combination of supervised machine learning models and unsupervised machine learning models that are in turn used to generate target features for a resultant prioritization machine learning model.

First claim

1 . A computer-implemented method for predictive prioritization of a plurality of predictive data entities, the computer-implemented method comprising:
identifying, by one or more processors, a plurality of feature data objects for the plurality of predictive data entities; generating, by the one or more processors, a refined multi-input-type supervised machine learning model based at least in part on one or more labeled feature data objects of the plurality of feature data objects that are associated with a labeled subset of the plurality of predictive data entities, wherein the refined multi-input-type supervised machine learning model is associated with one or more refined features; generating, by the one or more processors, a multi-dimensional unsupervised machine learning space that comprises a plurality of multi-dimensional entity mappings of the plurality of predictive data entities, wherein the multi-dimensional unsupervised machine learning space is associated with a plurality of mapping dimensions each associated with a refined feature of the one or more refined features; generating, by the one or more processors, a predicted prioritization score for each predictive data entity of the plurality of predictive data entities based at least in part on the multi-dimensional unsupervised machine learning space; and performing, by the one or more processors, one or more prediction-based actions based at least in part on each predicted prioritization score for a predictive data entity of the plurality of predictive data entities.

2 . The computer-implemented method of claim 1 , wherein identifying a feature data object of the plurality of feature data objects that is associated with a predictive data entity of the plurality of predictive data entities comprises:
determining one or more format-specific input data groupings based at least in part on a predictive entity data object for the predictive data entity; determining one or more updated format-specific input data groupings based at least in part on the one or more format-specific input data groupings and one or more external feedback data objects; performing an initial pattern recognition traversal of each updated format-specific input data grouping of the one or more updated format-specific input data groupings to detect a set of initial features for the updated format-specific input data grouping; performing one or more explanatory data analysis operations on the one or more updated format-specific input data groupings using the set of initial features to generate one or more updated features for the predictive entity data object; and generating the feature data object by processing the predictive entity data object in accordance with the one or more updated features.

3 . The computer-implemented method of claim 2 , wherein performing the one or more explanatory data analysis operations comprises:
generating a group of numerical representation values for the set of initial features in relation to the predictive data entity; detecting one or more co-linear features of the set of initial features based at least in part on a distribution of the group of numerical representation values across the plurality of predictive data entities; detecting one or more missing-value features of the set of initial features based at least in part on a count of missing values for the set of initial features in across the plurality of predictive data entities; and generating the one or more updated features for the predictive entity data object based at least in part on the one or more co-linear features and the one or more missing-value features.

4 . The computer-implemented method of claim 1 , wherein generating the refined multi-input-type supervised machine learning model comprises:
generating one or more initial multi-input-type supervised machine learning models based at least in part on the one or more labeled feature data objects; generating one or more refined features based at least in part on evaluating the one or more labeled data objects in relation to a prediction accuracy measure for each of the one or more initial multi-input-type supervised machine learning models; for each initial multi-input-type supervised machine learning model of the one or more initial multi-input-type supervised machine learning models, determining one or more model-related features of the one or more refined features; generating the one or more refined features based at least in part on each one or more model-related features for an initial multi-input-type supervised machine learning model; generating one or more model-related feature data objects based at least in part on the one or more refined features; generating an updated multi-input-type supervised machine learning model based at least in part on the one or more model-related feature data objects; generating one or more model testing outputs for the updated multi-input-type supervised machine learning model by testing the updated multi-input-type supervised machine learning model based at least in part on a subset of the one or more related data objects that fall outside a corresponding feature segment of the one or more labeled feature data objects for the updated multi-input-type supervised machine learning model; performing one or more hyper-parameter tuning operations for the updated multi-input-type supervised machine learning model based at least in part on the model testing outputs for the updated multi-input-type supervised machine learning model to generate a tuned updated multi-input-type supervised machine learning model; determining whether an accuracy metric for the tuned candidate trained multi-input-type supervised machine learning model is deemed optimal; and in response to determining that the accuracy metric for the tuned candidate trained multi-input-type supervised machine learning model is deemed optimal, generating the refined multi-input-type supervised machine learning model based at least in part on the tuned candidate trained multi-input-type supervised machine learning model.

5 . The computer-implemented method of claim 1 , wherein generating each predicted prioritization score for a predictive data entity of the plurality of predictive data entities comprises:
determining, based at least in part on the plurality of multi-dimensional entity mappings, one or more target entity clusters; determining, for each refined feature of the one or more refined features, a measure of correlation across the one or more target entity clusters; determining one or more target features of the one or more refined features based at least in part on each measure of correlation for a refined feature of the one or more refined features; generating a prioritization machine learning model based at least in part on the one or more target features; and processing each predictive data entity of the plurality of predictive data entities using the prioritization machine learning models to generate the predictive prioritization score for the predictive data entity.

6 . The computer-implemented method of claim 1 , wherein performing the one or more prediction-based actions is performed based at least in part on one or more recommended actions for each predictive data entity of the plurality of predictive data entities and a recommended action ordering of each one or more recommended actions for a predictive data entity of the plurality of predictive data entities.

7 . The computer-implemented method of claim 6 , wherein each one or more recommended actions for a predictive data entity of the plurality of predictive data entities describes one or more genetic testing actions for a patient profile of a plurality of patient profiles.

8 . The compute

Assignees

• Optum Inc

Inventors

• Naidoo Darrel
• Godden Paul J
• Boss Gregory J
• Wahl Peter M

Classifications

• G06N3/045

  Combinations of networks · CPC title

• G06N3/044

  Recurrent networks, e.g. Hopfield networks · CPC title

• G06N3/088 Primary

  Non-supervised learning, e.g. competitive learning · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/09

  Supervised learning · CPC title